Artemis: Anatomy-Resolved inTervention for Eliminating Multimodal NeuroImage confounderS

2026-06-10

AI Summary

This work addresses the challenge that demographic confounders such as age and sex in multimodal neuroimaging often lead graph neural networks (GNNs) to learn spurious associations rather than causally invariant representations. To mitigate this, the authors propose the first anatomy-aware, region-level causal intervention framework, which models confounding effects independently within each brain parcellation. By learning lightweight, region-specific confounder representations and integrating both functional (fMRI) and structural (DTI) connectivity features for graph-based reasoning, the method embeds clinically grounded causal mechanisms into the learning process, yielding neuroscientifically interpretable results. Evaluated on three major benchmarks—ADNI, OASIS, and HCP—the approach consistently outperforms existing GNN methods across tasks including disease diagnosis, dementia staging, and sex classification.

Abstract

Multimodal neuroimaging, integrating functional connectivity from fMRI and structural connectivity from DTI, enables non-invasive analysis of brain networks using graph neural networks. However, demographic factors such as age and sex systematically confound the relationship between brain connectivity and clinical outcomes, causing GNNs to exploit spurious shortcuts rather than learning causally invariant representations. While recent causal GNN methods introduce causality at the graph-modeling level, their causal mechanisms remain domain-agnostic without accounting for the real-world confounders inherent in clinical neuroimaging data. Moreover, brain networks are constructed from atlas-based parcellations where each region exhibits distinct sensitivity to demographic factors, necessitating region-aware adjustment. We propose Artemis, a region-level causal framework that bridges this gap with causal intervention at each brain region independently by learning region-specific confounder representations with lightweight parameters. Our adjustment comprehensively utilized the multimodal functional and structural features for graph reasoning as a plug-in module compatible with arbitrary GNN backbones. Experiments on three benchmarks, ADNI for disease diagnosis, OASIS for dementia staging, and HCP for sex classification, demonstrate consistent improvements over representative GNN-based baselines. Multiple supporting experiments further demonstrate statistical significance and neuroscientific interpretability.

Problem

• multimodal neuroimaging
• confounding factors
• graph neural networks
• causal representation
• brain parcellation

Innovation

• causal intervention
• region-aware adjustment
• multimodal neuroimaging
• graph neural networks
• confounder removal